Food supervision efficiency optimization method and system based on AI

Abstract

The invention relates to the technical field of food supervision, and discloses an AI-based food supervision efficiency optimization method and system. The method comprises the following steps: firstly, receiving a multi-source food supervision data stream in a target area, then analyzing data by using a risk feature extraction model to generate a risk level map and a probability matrix, then constructing an adaptive supervision strategy model to generate a supervision instruction set, and then simulating a risk linkage effect and optimizing parameters through a risk diffusion path prediction model. And finally, iteratively optimizing and outputting a supervision decision action sequence by using a multi-agent reinforcement learning framework. The system comprises a data receiving module, a processing analysis module, a strategy generation module, a risk simulation and parameter optimization module and a parameter iteration and decision output module. The method can accurately evaluate the risk, optimize the supervision strategy, predict the risk diffusion, improve the food supervision efficiency, and guarantee the food safety.

Description

Technical Field

[0001] The present invention relates to the field of food supervision technology, and specifically to an AI-based food supervision efficiency optimization method and system. Background Art

[0002] In today's society, food safety is of paramount importance. It directly impacts public health and safety, and has a profound impact on social stability and economic development. With the booming food industry, food production, distribution, and sales have become increasingly complex and diverse, and traditional food regulation approaches are facing numerous challenges.

[0003] From the perspective of food production, there are a large number of production enterprises with varying sizes. Under traditional regulatory methods, it is difficult for regulatory bodies to conduct high-frequency and detailed inspections on such a large number of enterprises, resulting in many safety issues in the production process being difficult to discover and deal with in a timely manner.

[0004] The food distribution process is also plagued by numerous problems. With the rapid development of e-commerce and the logistics industry, the scope and speed of food distribution are expanding. During transportation and storage, food is exposed to various environmental conditions, such as changes in temperature and humidity. However, under traditional regulatory models, the ability to monitor the trajectory of food logistics is limited, making it impossible to track changes in environmental parameters during food distribution in real time. If the temperature of food loses control during transportation due to a malfunction in cold chain equipment, regulators will have difficulty detecting it in a timely manner, and spoiled food may enter the market.

[0005] At the market end, traditional random inspection methods are somewhat unreliable. The selection of samples for random inspection often lacks scientific basis and may not accurately reflect the overall quality of food on the market. Furthermore, feedback on inspection results is delayed. By the time problematic food is discovered, a large number of products may have already reached consumers, making effective recall and disposal difficult. Furthermore, market conditions vary significantly across regions, making traditional regulatory methods difficult to tailor to local conditions and provide precise oversight.

[0006] Traditional regulatory approaches are particularly limited in terms of data processing and analysis capabilities. Faced with massive amounts of food regulatory data, such as production quality inspection records, logistics data from distribution channels, and market inspection reports, manual processing and analysis is not only inefficient but also prone to omissions. The inability to quickly and accurately identify potential food safety risks from this complex data leaves regulatory decisions lacking a scientific basis, making efficient food regulation difficult. Summary of the Invention

[0007] The purpose of the present invention is to provide an AI-based food supervision efficiency optimization method and system to solve the problems raised in the above background technology.

[0008] To achieve the above objectives, the present invention provides the following technical solution: an AI-based method for optimizing food supervision efficiency, the method comprising: Receive multi-source food regulatory data streams within the target area, including production quality inspection records, distribution logistics trajectory data, and market terminal random inspection reports; Based on a preset risk feature extraction model, the multi-source data is subjected to spatiotemporal normalization and feature correlation analysis to generate a food safety risk level map and a risk type probability matrix; Based on the risk level map, an adaptive supervision strategy model is constructed to generate a multi-dimensional supervision instruction set, which includes a sampling resource scheduling path and a dynamic deployment plan for supervision forces; Based on a preset risk diffusion path prediction model, simulate the chain reaction of food safety issues within a preset time period in the future and optimize the coordinated execution parameters of the multi-dimensional regulatory directive set; The collaborative execution parameters are iteratively optimized through a multi-agent reinforcement learning framework, and the regulatory decision action sequence is output to the food regulatory command platform.

[0009] Preferably, the steps of constructing the risk feature extraction model include: Collect a multi-year food safety incident case database to construct a multi-dimensional feature training set that includes production anomaly patterns, logistics disruption data, and risk trigger conditions; Performing feature dimensionality reduction on the multi-dimensional feature training set through a deep sparse autoencoder to extract independent representations of risk core factors and auxiliary factors; Combined with the food safety risk propagation dynamics equation, differential constraints on the dynamic correlation between factors are constructed; The differential constraint condition is embedded in a spatiotemporal graph convolutional network to generate the risk feature extraction model that supports online updating.

[0010] Preferably, the adaptive supervision strategy model includes: Dynamically classify the risk impact intensity levels according to the gradient distribution of the risk level map; Calculate risk coverage sensitivity scores based on regional population density and food supply chain vulnerability index; The sensitivity score and the impact intensity level are nonlinearly mapped using a sigmoid function to generate regional differentiated regulatory thresholds; The starting condition of the multi-level supervision protocol is triggered according to the threshold.

[0011] Preferably, the steps of constructing the risk diffusion path prediction model include: Collect chain reaction data from historical food safety incidents and construct a risk causal diagram dataset; Extract the transmission probability and delay parameters between risk events through causal reasoning algorithms; Combining complex network theory to construct a directed weighted graph of risk propagation and quantify the strength of dependencies between nodes; The dependency strength and the real-time regulatory intervention factor are input into the spatiotemporal attention network to generate the risk diffusion path prediction model.

[0012] Preferably, the method further comprises: identifying key conduction blocking nodes based on simulation results of the risk diffusion path prediction model; configuring a key monitoring strategy for the node in the multi-dimensional supervision instruction set; Based on the key monitoring strategy, a cross-regional collaborative supervision plan is automatically generated, including logistics channel blocking instructions and production-end traceability plans.

[0013] Preferably, the calculation of the risk coverage sensitivity score includes: Obtain real-time traffic hub flow matrix and detection resource distribution heat map to build regional risk resistance capacity assessment cube; Calculate high-order correlation weights between multi-dimensional features through hypergraph convolutional networks; Performing a tensor fusion operation on the association weight and the evaluation cube to obtain a comprehensive sensitivity score; The calculation formula for the comprehensive sensitivity score is: ; Where, Indicates the comprehensive sensitivity score value, Indicates the Vulnerability index of supply chain nodes, Indicates the The repair priority weight of the supply chain-like node, represents the regional basic risk resistance constant, represents the tensor Kronecker product, Represents the total number of supply chain node categories.

[0014] Preferably, the embedding of the differential constraint condition includes: Conduct dynamic stability analysis on the independent representations of the risk core factors and auxiliary factors to screen for correlation patterns that conform to physical laws; Generate feature evolution trajectories that meet the constraints through Markov chain Monte Carlo sampling; Trajectory data is used to regularize the edge weights of the spatiotemporal graph convolutional network to ensure that the model output conforms to the law of risk propagation.

[0015] Preferably, the execution of the multi-agent reinforcement learning framework includes: Define the reward function for multi-agent collaborative decision-making, including the dual objectives of risk control rate and resource allocation efficiency; The policy network for supervising agents in each region is trained through a phased curriculum learning strategy; In each round of training, the collaborative weights between agents are dynamically adjusted according to the degree of goal conflict; Outputting the regulatory decision action sequence that satisfies the equilibrium condition; Wherein, the reward function is: ; Where, Represents the reward value, represents the risk diffusion suppression score, represents the resource utilization efficiency score, is the dynamic balance factor.

[0016] Preferably, the method further comprises: After configuring the key monitoring strategy, the state transition probability of the conduction blocking node is monitored in real time; If the transfer probability exceeds the preset threshold, the simulated annealing optimization mechanism will be triggered to re-plan the spatiotemporal coordination plan for cross-regional supervision.

[0017] Preferably, the present invention also includes an AI-based food supervision efficiency optimization system, the system comprising: The data receiving module is used to receive multi-source food supervision data streams in the target area, including production quality inspection records, circulation logistics trajectory data, and market terminal random inspection reports; A data processing and analysis module is used to perform spatiotemporal normalization and feature correlation analysis on the multi-source data based on a preset risk feature extraction model to generate a food safety risk level map and a risk type probability matrix; A supervision strategy generation module is used to build an adaptive supervision strategy model based on the risk level map and generate a multi-dimensional supervision instruction set including a sampling resource scheduling path and a dynamic deployment plan for supervision forces; A risk simulation and parameter optimization module, which is used to simulate the chain reaction of food safety issues within a preset time period in the future based on a preset risk diffusion path prediction model, and optimize the coordinated execution parameters of the multi-dimensional regulatory instruction set; The parameter iteration and decision output module is used to iteratively optimize the collaborative execution parameters through a multi-agent reinforcement learning framework and output the regulatory decision action sequence to the food regulatory command platform.

[0018] Compared with the prior art, the present invention has the following beneficial effects: The present invention has achieved many significant beneficial effects in the field of food supervision through a series of innovative technical means. In terms of data processing and risk assessment, it receives multi-source food supervision data streams in the target area, including quality inspection records of the production link, logistics trajectory data of the circulation link, and market terminal sampling reports. The preset risk feature extraction model is used to perform spatiotemporal normalization and feature association analysis on these multi-source data, which can deeply explore the potential connections between the data. Compared with the traditional method of relying on manual data analysis, the accuracy and efficiency of risk assessment are greatly improved. For example, in the past, when manually analyzing data, some subtle but important risk associations may be ignored due to the huge amount of data and subjective factors of personnel. The model of the present invention can accurately capture this information, and the generated food safety risk level map and risk type probability matrix can provide comprehensive, intuitive and accurate risk status information to the regulatory body, so that regulators have a clearer understanding of food risks and make preparations in advance.

[0019] In the formulation of regulatory strategies, an adaptive regulatory strategy model is constructed based on the risk level map. By dynamically dividing the risk impact intensity level, combining the regional population density and the food supply chain vulnerability index to calculate the risk coverage sensitivity score, and then generating regional differentiated regulatory thresholds, triggering the start-up conditions of the multi-level regulatory agreement. This approach has changed the traditional "one-size-fits-all" regulatory model and achieved precise supervision. For high-risk areas and links susceptible to risks, more sampling resources and regulatory forces can be deployed in a timely manner to avoid waste of resources while ensuring that key areas and links are effectively supervised. Taking a certain city as an example, the same sampling frequency was used for different areas in the past, resulting in insufficient supervision of high-risk areas and waste of resources in low-risk areas. After applying the present invention, the sampling frequency of high-risk areas is increased according to the risk coverage sensitivity score, which greatly improves the detection rate of problematic foods. At the same time, the resource input in low-risk areas is reasonably reduced, optimizing resource allocation.

[0020] The ability to prevent and control risk diffusion has been significantly improved. Based on a preset risk diffusion path prediction model, the chain reaction of food safety issues within a preset time period in the future is simulated, and the collaborative execution parameters of the multi-dimensional regulatory instruction set are optimized. By predicting the risk diffusion path in advance, timely measures can be taken to block the spread of risks. For example, key transmission blocking nodes are identified, key monitoring strategies are configured for them, and cross-regional collaborative supervision plans are automatically generated, including logistics channel blocking instructions and production-end traceability plans. This means that when faced with food safety issues, the regulatory body no longer has to remedy the situation after the fact, but can take preventive measures in advance, reduce the scope of risk diffusion and the degree of harm, and effectively protect consumer health and market stability.

[0021] Leveraging a multi-agent reinforcement learning framework, collaborative execution parameters are iteratively optimized, outputting regulatory decision-making action sequences to the food regulatory command platform. This approach integrates the decisions of multiple regulatory agents, improving the scientific nature and coordination of decision-making. By defining a reward function that encompasses the dual objectives of risk control rate and resource allocation efficiency, a phased curriculum learning strategy is used to train the policy network of regulatory agents in each region. The collaborative weights between agents are dynamically adjusted based on the degree of conflict between the objectives, enabling regulatory decisions to achieve efficient resource utilization while controlling risks. In practical applications, this approach can reduce regulatory costs and improve regulatory efficiency while ensuring food safety, driving food regulatory work towards intelligent and efficient development. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 This is a diagram showing the working principle of the AI-based food supervision efficiency optimization method of the present invention; Figure 2 Schematic diagram of the process built for the adaptive supervision strategy model; Figure 3 Schematic diagram of the process executed for the multi-agent reinforcement learning framework; Figure 4 Schematic diagram of the process for optimizing key monitoring strategies and collaborative solutions. DETAILED DESCRIPTION

[0023] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0024] See also Figure 1-Figure 4 The present invention provides an AI-based food supervision efficiency optimization method and system, and its specific implementation method will be elaborated in detail below.

[0025] Within the target area, data collection equipment and network communication technologies are used to collect various food regulatory data in real time. Production quality inspection records include the test results of various quality indicators during the food production process, such as pesticide residue detection data for raw materials and hygienic indicators of the production environment. Logistics trajectory data during the distribution process uses IoT devices and logistics information systems to obtain records of changes in food's location, temperature, humidity, and other environmental parameters during transportation and storage. Market terminal random inspection reports, which are inspection reports obtained by regulatory bodies after food arrives at the sales terminal, cover information such as microbiological indicators and the use of additives.

[0026] The collected multi-source data is processed using a pre-defined risk signature extraction model. This model normalizes data across different temporal and spatial scales, making them comparable and consistent. Furthermore, the model's algorithms analyze the characteristics of each type of data, uncovering potential connections between them. This generates a food safety risk level map and a risk type probability matrix. The risk level map intuitively displays the risk level distribution of foods at different locations and stages within the target area; the risk type probability matrix clarifies the probability of different risk types occurring in various types of food.

[0027] Based on the generated risk level map, an adaptive regulatory strategy model is constructed. This model comprehensively considers multiple factors, such as the gradient distribution of risk levels, regional population density, and the food supply chain vulnerability index, to generate a multi-dimensional regulatory instruction set that includes a scheduling path for spot inspection resources and a dynamic deployment plan for regulatory forces. When determining the scheduling path for spot inspection resources, the order and frequency of spot inspection resource deployment are rationally arranged based on the risk level and regional characteristics. The dynamic deployment of regulatory forces is also considered in light of the distribution of risk areas and staffing, ensuring that regulatory forces can cover high-risk areas in a timely and effective manner.

[0028] Using a pre-defined risk diffusion path prediction model, we simulate the potential cascading effects of food safety issues within a pre-set timeframe. By analyzing historical data and the characteristics of the food supply chain, we predict the path and speed of risk transmission across different links and regions. Based on these simulation results, we optimize the collaborative execution parameters within the multi-dimensional regulatory directive set, ensuring closer coordination across regulatory links and improving regulatory efficiency.

[0029] A multi-agent reinforcement learning framework is used to iteratively optimize collaborative execution parameters. During this process, a reward function is defined that incorporates the dual objectives of risk control rate and resource allocation efficiency. The policy network of each regional supervisory agent is trained using a phased curriculum learning strategy. During each round of training, the coordination weights between agents are dynamically adjusted based on the degree of objective conflict. Ultimately, a supervisory decision-making action sequence that satisfies equilibrium conditions is output and transmitted to the food supervision command platform to provide decision support for supervisors.

[0030] The present invention will be further described below in conjunction with Examples 1 to 6: Example 1:

[0031] To build a risk feature extraction model, we collected a multi-year database of food safety incidents, covering safety incidents in different regions and for different food types. We extracted relevant data from these cases to construct a multidimensional feature training set encompassing production anomaly patterns, logistics disruption data, and risk trigger conditions. For example, production anomaly patterns might include abnormal fluctuations in parameters such as temperature and pressure during the production process; logistics disruption data might include information such as transportation downtime caused by vehicle failures and the duration of abnormal storage environments; and risk trigger conditions might include factors that may cause food safety issues, such as raw material contamination and processing violations.

[0032] Next, a deep sparse autoencoder is used to perform dimensionality reduction on the multidimensional feature training set. A deep sparse autoencoder is a deep learning model that automatically learns important features from data and removes redundant information. This process extracts independent representations of core risk factors and auxiliary risk factors. Core risk factors are factors that have a critical impact on food safety risks, such as the content of hazardous substances in food; auxiliary risk factors are factors that assist in risk assessment, such as the reputation of the manufacturer.

[0033] By combining the food safety risk propagation dynamics equation, we construct differential constraints on the dynamic relationship between factors. The food safety risk propagation dynamics equation describes the dynamic process of risk propagation in the food supply chain. By constructing differential constraints, we can more accurately characterize the interaction between the core risk factor and its auxiliary factors.

[0034] Differential constraints are embedded in a spatiotemporal graph convolutional network (STGCN) to generate a risk feature extraction model that supports online updates. During the embedding process, dynamic stability analysis is performed on the independent representations of the core and auxiliary risk factors to screen for correlation patterns that conform to physical laws. Markov chain Monte Carlo sampling is used to generate feature evolution trajectories that meet the constraints. This trajectory data is used to regularize the edge weights of the STGCN to ensure that the model output conforms to the laws of risk propagation. This training process enables the model to continuously adapt to new data, enabling online updates and improving the accuracy and timeliness of risk feature extraction.

[0035] Example 2: During the construction of the adaptive regulatory strategy model, the risk impact intensity levels are dynamically divided based on the gradient distribution of the risk level map. The risk level map is a visual representation of the food risk distribution, with its gradient distribution reflecting the trend of risk from high to low. By analyzing the gradient distribution, risk areas are divided into different intensity levels, such as high-risk, medium-risk, and low-risk areas.

[0036] Based on regional population density and the food supply chain vulnerability index, a risk coverage sensitivity score is calculated. A real-time transportation hub flow matrix and a heat map of testing resource distribution are obtained to construct a regional risk resilience assessment cube. The real-time transportation hub flow matrix reflects the flow of people and materials within the region, while the heat map of testing resource distribution shows the density of testing resources in different areas. The regional risk resilience assessment cube integrates this information to assess a region's risk resilience from multiple dimensions.

[0037] The hypergraph convolutional network (HGCN) calculates high-order correlation weights between multi-dimensional features. HGCNs are capable of processing complex relational data. In this scenario, they can analyze high-order correlations between multi-dimensional features such as regional population density, food supply chain vulnerability index, transportation hub flow, and detection resource distribution, and derive corresponding weights.

[0038] The association weights and the evaluation cube are subjected to tensor fusion operations to obtain the comprehensive sensitivity score. The calculation formula for the comprehensive sensitivity score is: ; Where, It represents the comprehensive sensitivity score value, which comprehensively reflects the sensitivity of the region to food safety risks. Indicates the The vulnerability index of a supply chain node is used to measure the vulnerability of different supply chain nodes when facing risks. The larger the value, the more susceptible the supply chain node is to risks. Indicates the The repair priority weight of the supply chain node class determines the priority of repairing different supply chain nodes after the risk occurs. It represents the constant of the basic risk resistance capacity of a region, and is a fixed value that reflects the basic conditions of the region's own risk resistance. Represents the tensor Kronecker product, which is used to perform operations at the tensor level and integrate information of different dimensions. It represents the total number of supply chain node classifications, which is determined based on the actual supply chain node classifications.

[0039] The sensitivity score and impact intensity level are nonlinearly mapped using a sigmoid function to generate regionally differentiated regulatory thresholds. The sigmoid function maps values from varying ranges to an appropriate interval, which then triggers the activation conditions of multi-level regulatory protocols, enabling precise regulation of different risk areas.

[0040] Example 3: When building the risk diffusion path prediction model, we collected chain reaction data from historical food safety incidents to construct a risk causal diagram dataset. This chain reaction data includes information such as the chronological relationship between risk events and the degree of impact. By organizing and analyzing this data, we constructed a risk causal diagram dataset for subsequent model training.

[0041] Causal inference algorithms are used to extract the transmission probability and delay parameters between risk events. Causal inference algorithms can uncover hidden causal relationships within data. In this process, they determine the transmission probability between different risk events—the likelihood that one risk event will trigger another. They also derive the delay parameters between risk events, reflecting the time interval between risk transmission.

[0042] Integrating complex network theory, we construct a directed weighted risk propagation graph to quantify the strength of dependencies between nodes. Complex network theory considers risk events as nodes in a network and risk propagation paths as edges. By constructing a directed weighted graph, we can intuitively demonstrate the propagation direction and dependency strength between risk events. The dependency strength between nodes is determined by factors such as transmission probability and delay parameters. A larger value indicates a closer connection between two nodes, and a greater likelihood of risk propagation and impact.

[0043] The dependency strength and real-time regulatory intervention factors are input into a spatiotemporal attention network to generate a risk diffusion path prediction model. The spatiotemporal attention network automatically focuses on key spatiotemporal information during the risk propagation process. By combining dependency strength with real-time regulatory intervention factors, such as spot checks and law enforcement measures taken by regulatory bodies, it more accurately predicts risk diffusion paths, providing strong support for regulatory decision-making.

[0044] Example 4: When making regulatory decisions based on the simulation results of the risk diffusion path prediction model, the first step is to identify key transmission blocking nodes. Key transmission blocking nodes are nodes in the risk transmission path that play a key role in preventing the further spread of risk. The location and attributes of these nodes are determined by analyzing the simulation results of the risk diffusion path prediction model.

[0045] The multi-dimensional regulatory directive sets configure key monitoring strategies for these nodes. These strategies include increasing the frequency of monitoring of key conduction-blocking nodes and improving the accuracy of monitoring indicators. For example, for a manufacturer at a critical link in the food supply chain (serving as a key conduction-blocking node), the frequency of testing at each stage, including raw material procurement, production, and finished product shipment, will be increased, while also increasing the accuracy requirements for testing items.

[0046] Based on key monitoring strategies, a cross-regional collaborative supervision plan is automatically generated, including logistics channel blocking instructions and production-end traceability solutions. When risk signs are detected at key transmission blocking nodes, a logistics channel blocking instruction is promptly issued based on the actual situation to prevent the continued circulation of potentially contaminated food. Simultaneously, the production-end traceability solution is activated to trace the food production source and identify the cause of the risk, so that targeted measures can be taken to address it and prevent further escalation of the risk.

[0047] After configuring the key monitoring strategy, the state transition probability of the conduction blocking node is monitored in real time. The state transition probability reflects the likelihood of a node transitioning from one state to another, such as from a normal production state to a state with potential risks. If the transition probability exceeds a preset threshold, the simulated annealing optimization mechanism is triggered to replan the spatiotemporal coordination plan for cross-regional supervision. The simulated annealing optimization mechanism is a heuristic optimization algorithm that can find a more optimal solution under certain conditions. By replanning the spatiotemporal coordination plan, the effectiveness of supervision can be further improved and food safety risks can be reduced.

[0048] Example 5: In the execution process of the multi-agent reinforcement learning framework, a reward function for multi-agent collaborative decision-making is defined, which includes the dual objectives of risk control rate and resource allocation efficiency. The reward function is: ; Where, Represents the reward value, which comprehensively measures the quality of multi-agent decision-making. It represents the risk diffusion suppression score, which is used to evaluate the inhibitory effect of decision-making on risk diffusion. The higher the score, the better the risk diffusion is controlled. It represents the resource utilization efficiency score, reflecting the efficiency of utilizing sampling resources, supervisory forces and other resources during the supervision process. The higher the score, the more reasonable the resource utilization. It is a dynamic balance factor, which can make a trade-off between risk control and resource allocation according to the actual situation. Its value range is arrive For example, during high-risk periods, When the risk is relatively low, you can reduce value and improve resource utilization efficiency.

[0049] The policy network for each regional supervisory agent is trained using a phased curriculum learning strategy. This strategy divides the training process into multiple phases, each with different learning objectives and tasks. Initially, the agent primarily learns basic supervisory strategies and rules. As training progresses, the learning difficulty and complexity are gradually increased, enabling the agent to adapt to different supervisory scenarios.

[0050] During each training round, the collaborative weights between agents are dynamically adjusted based on the degree of conflicting objectives. Since there may be some conflict between the two objectives of risk control rate and resource allocation efficiency, for example, greater risk control may require the investment of more resources, resulting in reduced resource utilization efficiency. By dynamically adjusting the collaborative weights, a balance can be found between the two objectives at different training stages and regulatory scenarios, improving the effectiveness of multi-agent collaborative decision-making. Ultimately, a regulatory action sequence that satisfies equilibrium conditions is output, providing scientific and rational decision-making recommendations for the food regulatory command platform.

[0051] Example 6: The AI-based food supervision efficiency optimization system of the present invention has a data receiving module responsible for receiving multi-source food supervision data streams within the target area, including production quality inspection records, circulation logistics trajectory data, and market terminal random inspection reports. This module achieves real-time and accurate data collection by connecting with various data acquisition devices and information systems. For example, it connects with the quality inspection equipment of the manufacturer to obtain production quality inspection records; communicates with the information management system of the logistics company to obtain logistics trajectory data; and interacts with the random inspection department database at the market terminal to obtain random inspection reports.

[0052] The data processing and analysis module, based on a pre-defined risk signature extraction model, performs spatiotemporal normalization and feature correlation analysis on multi-source data, generating a food safety risk level map and a risk type probability matrix. This module utilizes the risk signature extraction model and processes and analyzes the data according to the methods described in Example 1, providing data support for subsequent regulatory decisions.

[0053] The supervisory strategy generation module constructs an adaptive supervisory strategy model based on the risk level map, generating a multi-dimensional supervisory instruction set that includes a sampling resource scheduling path and a dynamic supervisory force deployment plan. This module builds the adaptive supervisory strategy model using the method described in Example 2. Combining the risk level map with other relevant factors, it develops a scientifically sound, multi-dimensional supervisory instruction set to ensure the effective allocation of supervisory resources.

[0054] The risk simulation and parameter optimization module, based on a pre-set risk diffusion path prediction model, simulates the cascading effects of food safety issues within a pre-set future timeframe and optimizes the coordinated execution parameters within the multi-dimensional regulatory directive set. This module uses the risk diffusion path prediction model and the method described in Example 3 to perform risk simulation. Based on the simulation results, it optimizes the coordinated execution parameters within the multi-dimensional regulatory directive set, improving the coordination and effectiveness of supervision.

[0055] The parameter iteration and decision output module iteratively optimizes collaborative execution parameters through a multi-agent reinforcement learning framework and outputs a regulatory decision action sequence to the food regulatory command platform. This module executes the multi-agent reinforcement learning framework according to the method described in Example 5, continuously optimizing the collaborative execution parameters. Ultimately, it transmits a regulatory decision action sequence that satisfies equilibrium conditions to the food regulatory command platform, providing supervisors with intuitive and actionable decision-making basis. Through the collaborative work of this series of modules, AI-based food regulatory efficiency optimization is achieved.

[0056] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "includes," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0057] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A food supervision efficiency optimization method based on AI, characterized in that: include: Receive multi-source food regulatory data streams within the target area, including production quality inspection records, circulation logistics trajectory data, and market terminal random inspection reports; Based on a preset risk feature extraction model, the multi-source food regulatory data stream is subjected to spatiotemporal normalization and feature correlation analysis to generate a food safety risk level map and a risk type probability matrix; Based on the risk level map, an adaptive supervision strategy model is constructed to generate a multi-dimensional supervision instruction set, which includes a sampling resource scheduling path and a dynamic deployment plan for supervision forces; Based on a preset risk diffusion path prediction model, simulate the chain reaction of food safety issues within a preset time period in the future and optimize the coordinated execution parameters of the multi-dimensional regulatory directive set; The collaborative execution parameters are iteratively optimized through a multi-agent reinforcement learning framework, and the regulatory decision action sequence is output to the food regulatory command platform.

2. The method for optimizing food supervision efficiency according to claim 1, characterized in that: The steps of constructing the risk feature extraction model include: Collect a multi-year food safety incident case database to construct a multi-dimensional feature training set that includes production anomaly patterns, logistics disruption data, and risk trigger conditions; Performing feature dimensionality reduction on the multi-dimensional feature training set through a deep sparse autoencoder to extract independent representations of risk core factors and auxiliary factors; Combined with the food safety risk propagation dynamics equation, differential constraints on the dynamic correlation between factors are constructed; The differential constraint condition is embedded in a spatiotemporal graph convolutional network to generate the risk feature extraction model that supports online updating.

3. The method for optimizing food supervision efficiency according to claim 1, characterized in that: The adaptive supervision strategy model includes: Dynamically classify the risk impact intensity levels according to the gradient distribution of the risk level map; Calculate risk coverage sensitivity scores based on regional population density and food supply chain vulnerability index; The sensitivity score and the impact intensity level are nonlinearly mapped using a sigmoid function to generate regional differentiated regulatory thresholds; The starting condition of the multi-level supervision protocol is triggered according to the threshold.

4. The method for optimizing food supervision efficiency according to claim 1, characterized in that: The steps of constructing the risk diffusion path prediction model include: Collect chain reaction data from historical food safety incidents and construct a risk causal diagram dataset; Extract the transmission probability and delay parameters between risk events through causal reasoning algorithms; Combining complex network theory to construct a directed weighted graph of risk propagation and quantify the strength of dependencies between nodes; The dependency strength and the real-time regulatory intervention factor are input into the spatiotemporal attention network to generate the risk diffusion path prediction model.

5. The method for optimizing food supervision efficiency according to claim 4, characterized in that: Also includes: identifying key conduction blocking nodes based on simulation results of the risk diffusion path prediction model; configuring a key monitoring strategy for the key conduction blocking node in the multi-dimensional regulatory instruction set; Based on the key monitoring strategy, a cross-regional collaborative supervision plan is automatically generated, including logistics channel blocking instructions and production-end traceability plans.

6. The method for optimizing food supervision efficiency according to claim 3, characterized in that: The calculation of the risk coverage sensitivity score includes: Obtain real-time traffic hub flow matrix and detection resource distribution heat map to build regional risk resistance capacity assessment cube; Calculate high-order correlation weights between multi-dimensional features through hypergraph convolutional networks; Performing a tensor fusion operation on the high-order association weight and the evaluation cube to obtain a comprehensive sensitivity score; The calculation formula for the comprehensive sensitivity score is: ； Where, Indicates the comprehensive sensitivity score value, Indicates the Vulnerability index of supply chain nodes, Indicates the The repair priority weight of the class node, represents the regional basic risk resistance constant, represents the tensor Kronecker product, Indicates the total number of node categories.

7. The method for optimizing food supervision efficiency according to claim 2, characterized in that: The embedding of the differential constraints includes: Conduct dynamic stability analysis on the independent representations of the risk core factors and auxiliary factors to screen for correlation patterns that conform to physical laws; Generate feature evolution trajectories that meet the constraints through Markov chain Monte Carlo sampling; Trajectory data is used to regularize the edge weights of the spatiotemporal graph convolutional network to ensure that the model output conforms to the law of risk propagation.

8. The method for optimizing food supervision efficiency according to claim 1, characterized in that: The implementation of the multi-agent reinforcement learning framework includes: Define the reward function for multi-agent collaborative decision-making, including the dual objectives of risk control rate and resource allocation efficiency; The policy network for supervising agents in each region is trained through a phased curriculum learning strategy; In each round of training, the collaborative weights between agents are dynamically adjusted according to the degree of goal conflict; Outputting the regulatory decision action sequence that satisfies the equilibrium condition; Wherein, the reward function is: ； Where, Represents the reward value, represents the risk diffusion suppression score, represents the resource utilization efficiency score, is the dynamic balance factor.

9. The method for optimizing food supervision efficiency according to claim 5, characterized in that: Also includes: After configuring the key monitoring strategy, the state transition probability of the conduction blocking node is monitored in real time; If the transfer probability exceeds the preset threshold, the simulated annealing optimization mechanism will be triggered to re-plan the spatiotemporal coordination plan for cross-regional supervision.

10. An AI-based food supervision efficiency optimization system, characterized by: include: The data receiving module is used to receive multi-source food supervision data streams in the target area, including production quality inspection records, circulation logistics trajectory data, and market terminal random inspection reports; A data processing and analysis module is used to perform spatiotemporal normalization and feature correlation analysis on the multi-source food regulatory data stream based on a preset risk feature extraction model to generate a food safety risk level map and a risk type probability matrix; A supervision strategy generation module is used to build an adaptive supervision strategy model based on the risk level map and generate a multi-dimensional supervision instruction set including a sampling resource scheduling path and a dynamic deployment plan for supervision forces; A risk simulation and parameter optimization module, which is used to simulate the chain reaction of food safety issues within a preset time period in the future based on a preset risk diffusion path prediction model, and optimize the coordinated execution parameters of the multi-dimensional regulatory instruction set; The parameter iteration and decision output module is used to iteratively optimize the collaborative execution parameters through a multi-agent reinforcement learning framework and output the regulatory decision action sequence to the food regulatory command platform.

Images